#!/usr/bin/env python # -*- encoding: utf-8 -*- ''' @File : pretrain.py @Time : 2025/04/10 16:43:43 @Author : Logan Zou @Version : 1.0 @Contact : loganzou0421@163.com @License : (C)Copyright 2017-2018, Liugroup-NLPR-CASIA @Desc : 基于 Transformers 的 LLM 预训练脚本 注：参考仓库：https://github.com/LlamaFamily/Llama-Chinese ''' import logging import math import os import sys from dataclasses import dataclass, field from torchdata.datapipes.iter import IterDataPipe, IterableWrapper from itertools import chain import deepspeed from typing import Optional,List import datasets import pandas as pd import evaluate import torch from datasets import load_dataset from datasets.combine import interleave_datasets import transformers from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR from transformers import ( CONFIG_MAPPING, MODEL_FOR_CAUSAL_LM_MAPPING, AutoConfig, AutoModelForCausalLM, AutoTokenizer, TrainerCallback, TrainerState, TrainerControl, HfArgumentParser, Trainer, TrainingArguments, default_data_collator, is_torch_tpu_available, set_seed, ) import datetime from transformers.testing_utils import CaptureLogger from transformers.trainer_utils import get_last_checkpoint from transformers.utils import check_min_version, send_example_telemetry from transformers.utils.versions import require_version from datasets import interleave_datasets logger = logging.getLogger(__name__) MODEL_CONFIG_CLASSES = list(MODEL_FOR_CAUSAL_LM_MAPPING.keys()) MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES) @dataclass class ModelArguments: """ Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch. """ model_name_or_path: Optional[str] = field( default=None, metadata={ "help": ( "The model checkpoint for weights initialization.Don't set if you want to train a model from scratch." ) }, ) model_type: Optional[str] = field( default=None, metadata={"help": "If training from scratch, pass a model type from the list: " + ", ".join(MODEL_TYPES)}, ) config_overrides: Optional[str] = field( default=None, metadata={ "help": ( "Override some existing default config settings when a model is trained from scratch. Example: " "n_embd=10,resid_pdrop=0.2,scale_attn_weights=false,summary_type=cls_index" ) }, ) config_name: Optional[str] = field( default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"} ) tokenizer_name: Optional[str] = field( default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"} ) cache_dir: Optional[str] = field( default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"}, ) use_fast_tokenizer: bool = field( default=True, metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."}, ) model_revision: str = field( default="main", metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."}, ) use_auth_token: bool = field( default=False, metadata={ "help": ( "Will use the token generated when running `huggingface-cli login` (necessary to use this script " "with private models)." ) }, ) torch_dtype: Optional[str] = field( default=None, metadata={ "help": ( "Override the default `torch.dtype` and load the model under this dtype. If `auto` is passed, the " "dtype will be automatically derived from the model's weights." ), "choices": ["auto", "bfloat16", "float16", "float32"], }, ) def __post_init__(self): if self.config_overrides is not None and (self.config_name is not None or self.model_name_or_path is not None): raise ValueError( "--config_overrides can't be used in combination with --config_name or --model_name_or_path" ) @dataclass class DataTrainingArguments: """ Arguments pertaining to what data we are going to input our model for training and eval. """ dataset_name: Optional[str] = field( default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."} ) dataset_config_name: Optional[str] = field( default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."} ) train_files: Optional[List[str]] = field(default=None, metadata={"help": "The input training data file (a text file)."}) validation_files: Optional[List[str]] = field( default=None, metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."}, ) max_train_samples: Optional[int] = field( default=None, metadata={ "help": ( "For debugging purposes or quicker training, truncate the number of training examples to this " "value if set." ) }, ) max_eval_samples: Optional[int] = field( default=None, metadata={ "help": ( "For debugging purposes or quicker training, truncate the number of evaluation examples to this " "value if set." ) }, ) streaming: bool = field(default=False, metadata={"help": "Enable streaming mode"}) block_size: Optional[int] = field( default=None, metadata={ "help": ( "Optional input sequence length after tokenization. " "The training dataset will be truncated in block of this size for training. " "Default to the model max input length for single sentence inputs (take into account special tokens)." ) }, ) overwrite_cache: bool = field( default=False, metadata={"help": "Overwrite the cached training and evaluation sets"} ) validation_split_percentage: Optional[int] = field( default=5, metadata={ "help": "The percentage of the train set used as validation set in case there's no validation split" }, ) preprocessing_num_workers: Optional[int] = field( default=None, metadata={"help": "The number of processes to use for the preprocessing."}, ) keep_linebreaks: bool = field( default=True, metadata={"help": "Whether to keep line breaks when using TXT files or not."} ) def __post_init__(self): if self.streaming: require_version("datasets>=2.0.0", "The streaming feature requires `datasets>=2.0.0`") if self.dataset_name is None and self.train_files is None and self.validation_files is None: raise ValueError("Need either a dataset name or a training/validation file.") else: if self.train_files is not None: extension = self.train_files[0].split(".")[-1] assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, a json or a txt file." if self.validation_files is not None: extension = self.validation_files[0].split(".")[-1] assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, a json or a txt file." def main(): # See all possible arguments in src/transformers/training_args.py # or by passing the --help flag to this script. # We now keep distinct sets of args, for a cleaner separation of concerns. parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments)) if len(sys.argv) == 2 and sys.argv[1].endswith(".json"): # If we pass only one argument to the script and it's the path to a json file, # let's parse it to get our arguments. model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1])) else: model_args, data_args, training_args = parser.parse_args_into_dataclasses() # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The # information sent is the one passed as arguments along with your Python/PyTorch versions. send_example_telemetry("run_clm", model_args, data_args) # Setup logging logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", handlers=[logging.StreamHandler(sys.stdout)], ) if training_args.should_log: # The default of training_args.log_level is passive, so we set log level at info here to have that default. transformers.utils.logging.set_verbosity_info() log_level = training_args.get_process_log_level() logger.setLevel(log_level) datasets.utils.logging.set_verbosity(log_level) transformers.utils.logging.set_verbosity(log_level) transformers.utils.logging.enable_default_handler() transformers.utils.logging.enable_explicit_format() # Log on each process the small summary: logger.warning( f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}" + f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}" ) logger.info(f"Training/evaluation parameters {training_args}") # Detecting last checkpoint. last_checkpoint = None if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir: last_checkpoint = get_last_checkpoint(training_args.output_dir) if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0: raise ValueError( f"Output directory ({training_args.output_dir}) already exists and is not empty. " "Use --overwrite_output_dir to overcome." ) elif last_checkpoint is not None and training_args.resume_from_checkpoint is None: logger.info( f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change " "the `--output_dir` or add `--overwrite_output_dir` to train from scratch." ) # Set seed before initializing model. set_seed(training_args.seed) # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below) # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/ # (the dataset will be downloaded automatically from the datasets Hub). # # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called # 'text' is found. You can easily tweak this behavior (see below). # # In distributed training, the load_dataset function guarantee that only one local process can concurrently # download the dataset. if True: data_files = {} dataset_args = {} if data_args.train_files is not None: print(data_args.train_files) data_files["train"] = data_args.train_files print('训练文件总个数',len(data_args.train_files)) if data_args.validation_files is not None: data_files["validation"] = data_args.validation_files extension = ( data_files["train"][0].split(".")[-1] if data_files["train"] is not None else data_args.validation_files.split(".")[-1] ) if extension == "txt": extension = "text" dataset_args["keep_linebreaks"] = data_args.keep_linebreaks raw_datasets = load_dataset( extension, data_files=data_files, streaming=data_args.streaming, cache_dir=os.path.join(training_args.output_dir,'dataset_cache'), use_auth_token=True if model_args.use_auth_token else None, **dataset_args, ) if data_args.streaming: raw_datasets = raw_datasets.shuffle(seed=training_args.seed, buffer_size=1000000) # If no validation data is there, validation_split_percentage will be used to divide the dataset. if "validation" not in raw_datasets.keys(): raw_datasets["validation"] = load_dataset( extension, data_files=data_files, split=f"train[:{data_args.validation_split_percentage}%]", cache_dir=model_args.cache_dir, use_auth_token=True if model_args.use_auth_token else None, **dataset_args, ) raw_datasets["train"] = load_dataset( extension, data_files=data_files, split=f"train[{data_args.validation_split_percentage}%:]", cache_dir=model_args.cache_dir, use_auth_token=True if model_args.use_auth_token else None, **dataset_args, ) # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at # https://huggingface.co/docs/datasets/loading_datasets.html. # Load pretrained model and tokenizer # # Distributed training: # The .from_pretrained methods guarantee that only one local process can concurrently # download model & vocab. config_kwargs = { "cache_dir": model_args.cache_dir, "revision": model_args.model_revision, "use_auth_token": True if model_args.use_auth_token else None, } if model_args.config_name: config = AutoConfig.from_pretrained(model_args.config_name, **config_kwargs) elif model_args.model_name_or_path: config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs) else: config = CONFIG_MAPPING[model_args.model_type]() logger.warning("You are instantiating a new config instance from scratch.") if model_args.config_overrides is not None: logger.info(f"Overriding config: {model_args.config_overrides}") config.update_from_string(model_args.config_overrides) logger.info(f"New config: {config}") print(training_args.local_rank,'start load tokenizer') tokenizer_kwargs = { "cache_dir": model_args.cache_dir, "use_fast": model_args.use_fast_tokenizer, "revision": model_args.model_revision, "use_auth_token": True if model_args.use_auth_token else None, } if model_args.tokenizer_name: tokenizer = AutoTokenizer.from_pretrained(model_args.tokenizer_name, **tokenizer_kwargs) elif model_args.model_name_or_path: tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path, **tokenizer_kwargs) else: raise ValueError( "You are instantiating a new tokenizer from scratch. This is not supported by this script." "You can do it from another script, save it, and load it from here, using --tokenizer_name." ) print(training_args.local_rank,'end load tokenizer') print(training_args.local_rank,'start load model') if model_args.model_name_or_path: torch_dtype = ( model_args.torch_dtype if model_args.torch_dtype in ["auto", None] else getattr(torch, model_args.torch_dtype) ) model = AutoModelForCausalLM.from_pretrained( model_args.model_name_or_path, from_tf=bool(".ckpt" in model_args.model_name_or_path), config=config, cache_dir=model_args.cache_dir, revision=model_args.model_revision, trust_remote_code=True, use_flash_attention_2=True, use_auth_token=True if model_args.use_auth_token else None, ) else: model = AutoModelForCausalLM.from_config(config,trust_remote_code=True) n_params = sum({p.data_ptr(): p.numel() for p in model.parameters()}.values()) logger.info(f"Training new model from scratch - Total size={n_params/2**20:.2f}M params") print(training_args.local_rank,'end load model') # We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch # on a small vocab and want a smaller embedding size, remove this test. embedding_size = model.get_input_embeddings().weight.shape[0] if len(tokenizer) > embedding_size: model.resize_token_embeddings(len(tokenizer)) # Preprocessing the datasets. # First we tokenize all the texts. if training_args.do_train: if data_args.streaming: dataset_head = raw_datasets["train"].take(3) print(list(dataset_head)) column_names = list(list(dataset_head)[0].keys()) else: column_names = list(raw_datasets["train"].features) else: if data_args.streaming: dataset_head = raw_datasets["validation"].take(3) column_names = list(list(dataset_head)[0].keys()) else: column_names = list(raw_datasets["validation"].features) print(column_names) text_column_name = "text" if "text" in column_names else column_names[0] # since this will be pickled to avoid _LazyModule error in Hasher force logger loading before tokenize_function tok_logger = transformers.utils.logging.get_logger("transformers.tokenization_utils_base") def tokenize_function(examples): with CaptureLogger(tok_logger) as cl: output = tokenizer( [ item for item in examples[text_column_name]]) return output with training_args.main_process_first(desc="dataset map tokenization"): if not data_args.streaming: tokenized_datasets = raw_datasets.map( tokenize_function, batched=True, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, load_from_cache_file=not data_args.overwrite_cache, desc="Running tokenizer on dataset", ) else: tokenized_datasets = raw_datasets.map( tokenize_function, batched=True, remove_columns=column_names, batch_size = 60000, ) if data_args.block_size is None: block_size = tokenizer.model_max_length if block_size > 1024: logger.warning( "The chosen tokenizer supports a `model_max_length` that is longer than the default `block_size` value" " of 1024. If you would like to use a longer `block_size` up to `tokenizer.model_max_length` you can" " override this default with `--block_size xxx`." ) block_size = 1024 else: if data_args.block_size > tokenizer.model_max_length: logger.warning( f"The block_size passed ({data_args.block_size}) is larger than the maximum length for the model" f"({tokenizer.model_max_length}). Using block_size={tokenizer.model_max_length}." ) block_size = min(data_args.block_size, tokenizer.model_max_length) # Main data processing function that will concatenate all texts from our dataset and generate chunks of block_size. def group_texts(examples): # Concatenate all texts. concatenated_examples = {k: list(chain(*examples[k])) for k in examples.keys()} # concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()} total_length = len(concatenated_examples[list(examples.keys())[0]]) # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can # customize this part to your needs. if total_length >= block_size: total_length = (total_length // block_size) * block_size # Split by chunks of max_len. result = { k: [t[i : i + block_size] for i in range(0, total_length, block_size)] for k, t in concatenated_examples.items() } # print(datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')) logger.info("group texts input examples length%d after_group size%d"%(len(examples['input_ids']),len(result["input_ids"]))) result["labels"] = result["input_ids"].copy() return result # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a remainder # for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value might be slower # to preprocess. # # To speed up this part, we use multiprocessing. See the documentation of the map method for more information: # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map with training_args.main_process_first(desc="grouping texts together"): if not data_args.streaming: lm_datasets = tokenized_datasets.map( group_texts, batched=True, num_proc=data_args.preprocessing_num_workers, load_from_cache_file=not data_args.overwrite_cache, desc=f"Grouping texts in chunks of {block_size}", batch_size = 40000, ) else: lm_datasets = tokenized_datasets.map( group_texts, batched=True, batch_size = 60000, ) print(training_args.local_rank,'start select train_dataset') if training_args.do_train: if "train" not in tokenized_datasets: raise ValueError("--do_train requires a train dataset") train_dataset = lm_datasets["train"] if data_args.max_train_samples is not None and data_args.streaming==False: max_train_samples = min(len(train_dataset), data_args.max_train_samples) train_dataset = train_dataset.select(range(max_train_samples)) print(training_args.local_rank,'end select train_dataset') if training_args.do_eval: if "validation" not in tokenized_datasets: raise ValueError("--do_eval requires a validation dataset") print(training_args.local_rank,'start select eval_dataset') eval_dataset = lm_datasets["validation"] if data_args.max_eval_samples is not None and data_args.streaming==False : max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples) eval_dataset = eval_dataset.select(range(max_eval_samples)) print(training_args.local_rank,'end select eval_dataset') def preprocess_logits_for_metrics(logits, labels): if isinstance(logits, tuple): # Depending on the model and config, logits may contain extra tensors, # like past_key_values, but logits always come first logits = logits[0] return logits.argmax(dim=-1) print(training_args.local_rank,'start load metric') metric = evaluate.load("accuracy.py") print(training_args.local_rank,'end load metric') def compute_metrics(eval_preds): preds, labels = eval_preds # preds have the same shape as the labels, after the argmax(-1) has been calculated # by preprocess_logits_for_metrics but we need to shift the labels labels = labels[:, 1:].reshape(-1) preds = preds[:, :-1].reshape(-1) return metric.compute(predictions=preds, references=labels) print(training_args.local_rank,'Initialize our Trainer') trainer = Trainer( model=model, args=training_args, train_dataset= IterableWrapper(train_dataset) if training_args.do_train else None, eval_dataset= IterableWrapper(eval_dataset) if training_args.do_eval else None, tokenizer=tokenizer, # Data collator will default to DataCollatorWithPadding, so we change it. data_collator=default_data_collator, compute_metrics=compute_metrics if training_args.do_eval and not is_torch_tpu_available() else None, preprocess_logits_for_metrics=preprocess_logits_for_metrics if training_args.do_eval and not is_torch_tpu_available()else None, # callbacks=([SavePeftModelCallback] if isinstance(model, PeftModel) else None), ) if training_args.do_train: checkpoint = None if training_args.resume_from_checkpoint is not None: checkpoint = training_args.resume_from_checkpoint elif last_checkpoint is not None: checkpoint = last_checkpoint print(training_args.local_rank,'start train') train_result = trainer.train(resume_from_checkpoint=checkpoint) trainer.save_model() # Saves the tokenizer too for easy upload metrics = train_result.metrics max_train_samples = ( data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset) ) metrics["train_samples"] = min(max_train_samples, len(train_dataset)) trainer.log_metrics("train", metrics) trainer.save_metrics("train", metrics) trainer.save_state() # Evaluation if training_args.do_eval: logger.info("*** Evaluate ***") metrics = trainer.evaluate() max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset) metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset)) try: perplexity = math.exp(metrics["eval_loss"]) except OverflowError: perplexity = float("inf") metrics["perplexity"] = perplexity trainer.log_metrics("eval", metrics) trainer.save_metrics("eval", metrics) def _mp_fn(index): # For xla_spawn (TPUs) main() if __name__ == "__main__": main()